Design of integrated neural network model for weld seam tracking and penetration monitoring

作者：Ding, Dukun^*

来源：Cluster Computing, 2017, 20(4): 3345-3355.

DOI：10.1007/s10586-017-1084-0

• Weld pool centroid
• Measurement and control integration model
• Modelling
• Neural network

This paper has proposed a new method for welding procession monitoring and controlling. A Cartesian welding experimental platform is firstly setup. Then welding experiments can be performed and several groups of welding images can be captured by the vision sensor. A composite filtering system is set up and used to filter the weld arc disturbance. Then median filter and grey level transformation operations are performed to enhance the contrast of processing district on the image. On this basis,the grey centroid of weld pool C, the width of outer weld pool W, the width of inner weld pool N and welding current I are seem to be the heat distribution condition parameters. The weld seam error e and the welding penetertion condition p are used to be the welding conditon detection parameters. A BP neural network is used to set up the relationship between the heat distribution condition parameters and the welding condition detection parameters. In the end, several testing experiments are performed. The resluts show that the pediction value of the tracking error are fit to the measuring value, and the average tracking error is 0.011 mm. The accuracy rate of the model for welding penetration prediction is up to 95%.

• 出版日期2017-12
• 单位东莞理工学院